Embed Size (px)
Citation preview
NBER WORKING PAPER SERIES
THE ACQUISITION AND COMMERCIALIZATION OF INVENTION IN AMERICAN MANUFACTURING:INCIDENCE AND IMPACT
Ashish AroraWesley M. Cohen
John P. Walsh
Working Paper 20264http://www.nber.org/papers/w20264
NATIONAL BUREAU OF ECONOMIC RESEARCH1050 Massachusetts Avenue
Cambridge, MA 02138June 2014
This research was funded by grants from the National Science Foundation, and the Kauffman Foundation.Additional support was provided by The Fuqua School of Business, Duke University, The Schoolof Public Policy, Georgia Tech and the National Institutes of Health. You-Na Lee and Colleen Cunninghamprovided outstanding research assistance. We are also grateful to Vaibhav Gajulapalli and Arun Patrofor helping with data construction, and to seminar participants at Carnegie Mellon, Columbia, Duke,Berkeley, Michigan, New York University, Toronto, University of North Carolina (Greensboro), Tilburg,KU Leuven, Cambridge, London Business School, Japan’s Research Institute of Economy, Tradeand Industry, the Roundtable for Engineering Entrepreneurship Research at Georgia Tech, the AtlantaCompetitive Advantage Conference, the International Joseph A. Schumpeter Conference held in Jena,Germany and the IP2 Conference at the Hoover Institution for helpful comments. We are also gratefulto two anonymous reviewers of this journal for many helpful comments and suggestions. The customarydisclaimers apply. The views expressed herein are those of the authors and do not necessarily reflectthe views of the National Bureau of Economic Research.
NBER working papers are circulated for discussion and comment purposes. They have not been peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies officialNBER publications.
© 2014 by Ashish Arora, Wesley M. Cohen, and John P. Walsh. All rights reserved. Short sectionsof text, not to exceed two paragraphs, may be quoted without explicit permission provided that fullcredit, including © notice, is given to the source.
The Acquisition and Commercialization of Invention in American Manufacturing: Incidenceand ImpactAshish Arora, Wesley M. Cohen, and John P. WalshNBER Working Paper No. 20264June 2014, Revised September 2015JEL No. L1,O3,O30,O31,O32,O34
ABSTRACT
Recent accounts suggest the development and commercialization of invention has become more “open.”Greater division of labor between inventors and innovators can enhance social welfare through gainsfrom trade and greater economies of specialization. Moreover, this extensive reliance upon outsidesources for invention also suggests that understanding the factors that condition the extramural supplyof inventions to innovators is crucial to understanding the determinants of the rate and direction ofinnovative activity.
This paper reports on a recent survey of over 6000 American manufacturing and service sector firmson the extent to which innovators rely upon external sources of invention. Our results indicate that,between 2007 and 2009, 16% of manufacturing firms had innovated – meaning had introduced a productthat was new to the industry. Of these, 49% report that their most important new product had originatedfrom an outside source, notably customers, suppliers and technology specialists (i.e., universities, independentinventors and R&D contractors). We also estimate the contribution of each source to innovation inthe US economy. Although customers are the most frequent outside source, inventions acquired fromtechnology specialists tend to be the more economically more significant in term of their gross commercialvalue. As a group, external sources of invention make a significant contribution to the overall rateof innovation in the economy. Innovation policies, both public and private, should pay careful attentionto external supply of invention, and the efficiency of the mechanisms mediating between inventorsand innovators.
Ashish AroraFuqua School of BusinessDuke UniversityBox 90120Durham, NC 27708-0120and [email protected]
Wesley M. CohenThe Fuqua School of BusinessDuke UniversityBox 90120Durham, NC 27708-0120and [email protected]
John P. WalshGeorgia Institute of TechnologySchool of Public [email protected]
2
1. Introduction Until recently, the dominant model of innovation conceived of the innovation process as
typically carried out within the confines of a given firm, starting with the firm investing in R&D
to generate inventions, and then developing and commercializing those inventions. This model,
which reflected the large scale investments in internal R&D made by U.S. firms from the 1940s
through the 1970s, has become less accurate as a description of the innovation process over time
(Chesbrough, 2003). But we lack systematic evidence on the extent to which innovation
introduced by US manufacturers relies upon external invention..
Distinguishing between invention (i.e., discovery) and innovation (i.e., the
commercialization of an invention), this paper presents results from a survey of over 6000
American manufacturing and service sector firms on the extent to which innovating firms obtain
inventions from external sources. We first identify innovating firms, and then document the
share of innovating firms that acquire their inventions from different sources as well as the
channels through which firms acquire those inventions. We also characterize how the
distribution of value from invention compares across different sources. Finally, we estimate the
relationship between external sourcing of inventions and the innovative performance of the
manufacturing sector.
Why is our study of what Arora and Gambardella (1994) call the “division of innovative
labor” important? The availability and use of external sources of invention may offer important
social welfare benefits. First, there are social welfare gains from trade. When the firms best
equipped to invent are not necessarily the firms most capable of commercializing invention,
society benefits when an invention can transfer between them. Economic theory, starting with
Adam Smith, further suggests that such a division of innovative labor should also confer system-
wide efficiencies through economies of specialization.
In addition to the welfare implications of a division of innovative labor, our study offers
implications for empirical scholars of technological change. For example, if reliance upon
external sources of invention is pervasive, then an understanding of the factors affecting the
availability of those inventions should help us understand an industry’s rate of innovation.
Moreover, to the degree that other firms in other industries are supplying these inventions, an
understanding of an industry’s innovative activity will need to extend beyond conventionally
considered determinants of innovation applicable to the innovating firms’ own industry (e.g.,
3
appropriability, market demand, and technological opportunity) to those applicable to the
markets of the potential suppliers of invention.
A number of related literatures have already examined the links across firms and between
firms and other institutions that contribute to innovative performance. One literature has focused
on knowledge spillovers across rivals, as distinct from the purposeful sourcing of invention (e.g.,
Griliches, 1992). Other studies have focused on particular sources of inventions—including
startups (e.g., Arora and Gambardella, 1990), universities (e.g., Klevorick et al., 1995;
Henderson et al., 1998; Cohen et al., 2002), suppliers (Pavitt, 1984), and customers (e.g. von
Hippel, 1986)—as important contributors to the innovations produced by firms.
Building upon Mueller (1962), Jewkes, Sawers and Stillerman (1969), Freeman (1991)
and others, survey-based studies of the American manufacturing sector have considered a broad
set of knowledge sources for innovative activities, such as buyers, suppliers and universities
(Klevorick et al., 1995; Cohen et al., 2002a, 2002b). More recently, scholars using data gathered
through the Community Innovation Survey (CIS) have also reported findings on the different
sources contributing to firms’ innovative activities in Europe. Examples of these studies include
Tether (2002) for the UK, Veugelers and Cassiman (1999) for Belgium, Lhuillery, S., & Pfister,
E. (2009) for France, and Belderbos et al (2004) for Netherlands. Mohnen and Roeller (2005)
and Tether and Tajar (2008) are examples of studies that use cross-country CIS data. These
studies tend to find that external collaborations are important for innovation. Moreover,
customers tend to be the most common source of knowledge contributing to firms’ innovative
activities.
Other studies have examined specific channels through which firms’ benefit from
extramural sources of knowledge, such as licensing (e.g., Arora and Gambardella, 1990; Arora et
al., 2001) and collaboration (Hagedoorn, 2003). More recently, capitalizing on successive
versions of the CIS, others have focused on the contribution of collaboration to firms’ innovative
activities (e.g., Tether, 2002; Laursen and Salter, 2006, 2014; Grimpe and Sofka, 2009, Sofka
and Grimpe, 2010, Belderbos et al. 2004).
Our paper offers several advances. First, we report on the contribution of external sources
to firms’ innovative activities in the U.S., as distinct from prior studies for the U.S. that focused
on the contributions of external sources to R&D per se (Klevorick et al., 1995; Cohen et al.,
2002a, 2002b). Second, our survey is the first to focus on whether an innovating firm obtains the
4
underlying invention from an outside source, unlike prior work that examined the importance of
information flows from different sources or collaboration with external sources.
Unlike the prior literature, our survey focuses on the respondent’s most important product
innovation rather than the respondents’ innovations in general. Inquiring about a specific
innovation allows us to link the source of the invention to the channel employed, as well as to
other activities and outcomes, such as the percentage of sales due to the specific innovation, and
whether the innovation has a patent. We are also able to tie the specific innovation to activities to
commercialize the innovation, including investment in equipment, hiring of new personnel, or
the development of new distribution channels. Our data also permit us to compare the relative
value of inventions originating from different sources by allowing us to distinguish between the
cost and benefits of different sources by providing measures of the frequency with which
inventions are obtained from a particular source that we then combine with various indicators of
value, such as the share of sales from the innovative product, whether it was patented, and
whether the innovator made complementary investments.
To preview our findings, we find that, of the 16% of the U.S. manufacturing firms that
innovated (i.e., had introduced a new or significantly improved product to the industry) between
2007 and 2009, 49% report that their most important new product originated from an identified
outside source, suggesting pervasive reliance upon external sources of invention. Just over a
third of these transfers involves a market transaction, such as a license, a service contract or an
equity purchase. We also find that, although customers are the most pervasive source of outside
invention, those originating from what we call “technology specialists”—independent inventors,
contractors and universities—tend to be more valuable. Finally using two different analytic
approaches, we conclude that denying firms the opportunity of going to an outside source could
lead to a reduction of as much as 33% to 45% in the share of innovating firms in the U.S.
manufacturing sector.
To motivate the study, section 2 offers a simple model to illustrate how external supply of
inventions may affect innovation. In section 3, we describe the survey design underlying our
sample. Section 4 presents our findings with respect to innovation rates, the sources of
innovation, and the channels through which innovators acquire the inventions underlying their
innovations. We also compare the relative cost and value of the inventions originating from
different sources, and analyze the relationship between innovation rates and the extramural
5
sourcing of invention. Section 5 summarizes our main findings, discusses the implications and
concludes.
2. Background
If external sourcing of invention is pervasive, this has important implications for our
understanding of the drivers of innovation. To motivate our empirical analysis, we offer a simple
illustrative model to show that external sources do not simply provide a choice between internal
R&D and the purchase of an invention from an outside source. That is, they do not only affect a
firm’s “make or buy” decision.4 External availability of invention can also affect whether firms
decide to innovate in the first instance.
In Figure 1, we assume that the firm has a demand schedule for invention derived from
product market demand. We also assume it faces an upward marginal cost schedule for
invention, which represents the cost of generating innovation internally. Without external supply,
the equilibrium quantity of inventions is Q2. We further assume, however, that the firm can
access external inventions at a constant cost, w. Figure 1 allows us to make two points. First, it
shows that the availability of externally sourced invention can increase the overall rate of
invention, and, in turn, innovation from what it would be in the absence of external supply, in
this case from Q2 to Q3. Second, the presence of external sources of invention also yields some
substitution of external inventions for internal inventions, represented by Q2 - Q1. Thus, we
expect the supply of external technology to affect both the overall rate of innovation and the
share of internally generated innovations.
The cost of externally available inventions, w, does not simply reflect the cost of their
generation by external parties. There are also numerous costs associated with the acquisition of
invention from external sources. The relevant information and rights must be transmissible
across firms, and thus be applicable outside the context in which it was developed (von Hippel
1990; Arora and Gambardella, 1994). Once transmitted, the knowledge is at risk of
misappropriation, and patents in practice offer effective protection in only a small number of
industries (Scherer et al., 1959; Levin et al., 1987; Cohen et al., 2000). Williamson (1991) and
4 For instance, Pisano (1990), in his study of pharmaceutical industry R&D, is one of many who consider the issue
of relying on external invention strictly in terms of “make-or-buy.”
6
Figure 1: Supply and Demand for Inventions
Teece (1986) also highlight the role of transaction costs in limiting market transactions in
knowledge. Mowery (1983) and Kline and Rosenberg (1986) further note the need for ongoing
coordination and mutual adjustment across different innovation stages that can further impede
the writing of complete contracts. Finally, it is costly for buyers and sellers of inventions to find
one another. Thus the transfer of inventions across entities is often fraught with difficulties,
imposing costs beyond simply the price of the invention (cf. Arora and Gambardella, 2010).
3. Data: Survey design Our data are from a phone survey of firms in U.S. manufacturing and selected service
sector industries (see Table A1 in the appendix for a list of the industries in the sample). 5
In this
paper we focus on the manufacturing sector alone. Our sampling frame was the Dun and
Bradstreet (D&B) Selectory database. Note that we are not sampling on innovators (as done in
studies based on patent data), nor on R&D performers (as done in the Carnegie Mellon Survey
(Cohen et al., 2000).6
To obtain a substantial number of innovators from each industry, we stratified our sample
along multiple dimensions. Because all cases stay in the sample, errors in the D&B data used for
5 NORC, at the University of Chicago, administered our survey.
6 This sampling strategy is analogous to that employed by the Community Innovation Survey (CIS) in Europe and
the U.S. National Science Foundation’s Business R&D and Innovation Survey (BRDIS).
7
stratification only affect the efficiency of the sampling, not its representativeness (Kalton, 1983).
To begin, we selected all the D&B cases in our population of industries. We took all the Fortune
500 firms in our sample and collected information on all the subsidiaries of those firms that were
in our population industries, even if those were not the main industry of the parent firm. These
subsidiaries were grouped into business units, defined as a firm’s activities within a given
NAICS, with each subsidiary grouped by its primary NAICS. For these Fortune 500 firms, the
sampling unit is the firm’s activity in a NAICS (so that a diversified Fortune 500 firm may
appear multiple times in the sample). All other firms were assigned a single sampling unit based
on their primary NAICS. The sample was stratified into 28 industries, at the 3 or 4 digit NAICS.
Finally, the sampling frame was divided by size (Fortune 500, over 1000 employees but not
Fortune 500, 500 to 1000 employees, 100 to 499 employees and 10 to 99 employees, and less
than 10 employees), and by startup or not (less than five years old versus five or more years old).
We oversampled large firms (Fortune 500, which were sampled with certainty across all
business units, and firms of over 1000 employees), startup firms, those from more innovative
industries (using Community Innovation Survey (CIS) data from Europe to estimate innovation
rates for each industry), those in NAICS 533 as a primary or secondary industry (lessors of
intellectual property) and less populated industries (to ensure minimum sample sizes for
industry-level estimates). Other categories were under-sampled. We use a post-sample weighting
procedure (described below) to make the D&B data representative of U.S. manufacturing
industries.
The survey design included cognitive testing of the questionnaire against potential
respondents, pre-testing of the instrument and protocol, and multiple rounds of follow-up
contacts to increase response rate. The survey instrument was designed with a branching logic so
that non-innovative firms received only a brief questionnaire, and firms that innovated were
asked more details about their innovation process and outcomes. The sample consisted of 28,709
cases. An initial screening eliminated many cases, (for example, bakeries that are in retail, not
manufacturing) leaving a final sample of 22,034. The interview protocol started with a D&B
8
contact name (ideally the marketing manager or, for smaller firms, the business manager), and
then worked through the receptionist or other contacts to find an appropriate respondent.7
The survey was in the field from May to October, 2010. In the end, we received 6685
responses, yielding an adjusted 30.3% response rate. Appendix Table A2 shows the response
rates by industry and firm size. Non-response bias tests comparing D&B data for respondents
and non-respondents show that the sample represents the population on firm age, being
multiproduct, region, or likelihood to export (how international the firm is). Units of Fortune 500
firms were somewhat less likely to respond (about 20% response rate). Similarly, large firms,
multi-unit firms and public firms are somewhat less likely to respond. With regard to industry
responsiveness, pharmaceuticals had a low response rate, but still over 20%. Further cleaning
(based on recoding industries according to survey responses rather than initial D&B
categorizations) identified another 179 out-of-population respondents. For the remaining sample,
D&B business unit industry classifications were confirmed, and if necessary updated, based on
survey responses. In addition, for the purposes of this paper, we exclude the very smallest
establishments (less than 10 employees) and non-manufacturing establishments. The result is a
sample of 6088 cases. Appendix Table A3 presents the distribution of our sample based on
NAICS (disaggregated to 48 industries with at least 30 cases in each industry) and firm size
(collapsed into small [less than 100 employees], medium [100-1000 employees] and large [over
1000 employees]).
We reweighted the sample with post-sample weights based on Census data on the
population of firms in our industries, size strata and age. We constructed a matrix of these three
dimensions of stratification from a custom report provided by the U.S. Bureau of the Census,8
and then constructed a set of weights for our 6088 responses that reflect the population
distribution on this three dimensional matrix. After applying these weights, our sample should
represent the underlying population in terms of the industry-size-startup distribution (Kalton,
1984). These weights are used in all the empirical results in this paper.
7 According to the interview script, an appropriate respondent would be “the marketing manager or another person
in your company familiar with the firm’s products and services.” This flexibility in finding an appropriate contact
person was a key rationale for using a phone survey over post-mail or email surveys. 8 We thank Ron Jarmin and his team at the U.S. Bureau of the Census for providing this report.
9
4. Findings
4.1 Innovation
In this study, we focus exclusively upon product, rather than process, innovations. Following
prior innovation surveys, we asked the respondent if the firm had earned any revenue in 2009
from a new or significantly improved (NOSI) product or service introduced since 2007. For
those that said yes, we then asked whether their most significant innovation -- defined as that
product innovation accounting for the plurality of 2009 sales in the respondent’s market -- was
new to the market (NTM), i.e. introduced “in this industry before any other company”.9 We do
not specify a geographical boundary to the “industry” (for example, we are not limiting to a local
or domestic “market”).10
Table 1 provides illustrative examples of innovations introduced by
firms in the manufacturing sector. We also asked whether our respondents had developed
technology for other firms.
About 10% of our respondents report that they had both supplied technology to another
firm and also introduced new products to the market. Only 4% of manufacturing firms had
developed technology for another firm but had not themselves introduced a new product to the
market. These firms may be considered specialized technology suppliers. In this paper, we shall
focus on firms that commercialize new products.
Table 2 presents summary statistics for the rates of innovation overall and by industry,
where we aggregate our observations of firms in the manufacturing sector into 17 industry
groups, defined largely at the 3-digit NAICs level. The figures in Table 2 and all subsequent
tables are weighted to be representative of firm size (when reported at the industry level),
industry (when reported by firm size) and the true firm size and industry distribution (when
reported at the aggregate level).
Table 2 shows that 42% of firms report introducing a new-to-the-firm (not necessarily
new to the market) or significantly improved (NOSI) product in the prior three years. We also
find significant differences in the rates of new or improved product introduction across
9 Our NTM figure may underestimate the percentage of firms introducing NTM innovations. For example, a firm’s
most significant (i.e., highest selling) innovation may not be NTM, but its second most significant is, implying that
the firm is incorrectly classified as a not an “NTM” innovator. However, any bias is likely to be small because a
sizeable fraction of firms introduce only at most one innovation during the sample period, as discussed below. 10
We also did not count as innovators firms that reported either that they introduced their “most significant
innovation” outside of the 2007-2009 time window, or reported zero 2009 sales revenue due to this innovation.
10
industries. For example, at least 60% of firms in electronics, pharmaceuticals and
semiconductors introduced a NOSI product, while barely one-third of firms in wood or mineral
products did so. If we limit product innovations to the introduction of something new to the
market (NTM), we find that 16% of manufacturing firms have introduced such an innovation,
with substantially higher rates in instruments, electronics and pharmaceuticals, while wood and
metals have lower rates. Thus, 38% of firms with a NOSI product are the first to bring that
product to market, and are classified as innovators. In what follows, the term “innovation” refers
to products that are new to the market.
Table 1: Examples of innovations in sampled industries.
Table 2 also shows that larger firms are more likely to innovate and more likely to
introduce new products. We find that 39% of small firms but 65% of large firms introduce new
NOSI products. For innovations (i.e., new to the market products), the rates were 13% and 38%
Industry Innovation Food Antioxidant chocolates
Food Live active cheddar cheese with probiotics
Beverage vitamins enhanced flavoured spring water
Textile Heat resistant yarn
Textile New varieties of garments
Paper Low surface energy light tapes resistant to air, water, detergents, moisture, UV light, and dust
Paper Hanging folder with easy slide tab
Petroleum Non detergent motor oil
Chemicals BioSolvents – water based emulsion technology
Pharmaceutical Oral gallium to prevent bone decay
Pharmaceutical inhalation anaesthetics
Plastics Styrene based floor underlayment
Minerals Multi-wall polycarbonate recyclable panels
Minerals Solar glass and coating technologies solar modules
Metals Solder system & nanofoils
Metals New water faucets and bath products
Electronics USB-to-GPIB Interface Adapter
Electronics 20-h IPS Alpha LCD Panel
Semiconductors Linear voltage regulators
Semiconductors Phase change memory
Transport Equipment Improved alcohol sensing system
11
for small and large firms, respectively. Thus, larger firms are more likely to have at least one
innovation. This result is expected since the respondent is reporting if there is at least one
innovation, which should be more likely for larger firms (cf. Cohen and Klepper, 1996). If we
interpret the difference between NOSI and NTM as measuring imitation, it follows that
innovation increases by firm size but imitation is relatively stable across firm size classes. A
similar stability in imitation rates is also observed across industries.
Table 2. Rates of innovation and imitation, patenting and % sales for U.S. mfg. industries.
INDUSTRY (Number of respondents)
% NOSI . a
% NTM . b
%Imitator .
a-b
% sales from NOSI
% sales from focal NTM innovation
% NTM patented
Food/Bev (362) 40% 13% 27% 16% 9% 24% Textiles (210) 37% 15% 22% 19% 15% 51% Wood (385) 33% 8% 25% 15% 7% 11% Chemicals (365) 49% 24% 25% 17% 9% 42% Pharma (128) 62% 28% 33% 23% 13% 61% Plastics (340) 47% 16% 31% 14% 6% 42% Minerals (323) 30% 9% 21% 21% 14% 35% Metals (324) 38% 9% 29% 14% 5% 23% Fab Metals (424) 38% 10% 28% 28% 8% 35% Machinery (384) 44% 20% 24% 24% 14% 52% Electronics (146) 76% 33% 43% 38% 9% 58% Semicond (302) 60% 27% 33% 29% 18% 59% Instruments (135) 59% 37% 22% 17% 7% 54% Elec Equip (344) 54% 26% 28% 25% 13% 53% Auto (339) 50% 27% 23% 25% 11% 34% Med Equip (136) 55% 22% 33% 37% 31% 72% Misc. (510) 47% 19% 29% 30% 10% 45% All manuf. (5157) 42% 16% 27% 22% 11% 42% Large firms (1268) 65% 38% 27% 24% 10% 63% Med. firms(945) 54% 23% 31% 20% 15% 47% Small firms (2944) 39% 13% 26% 19% 12% 36%
For those respondents reporting that they innovated (i.e., introduced a NTM innovation),
the fifth and sixth columns of Table 2 show the percentage of 2009 sales represented,
respectively, by all new-to-the firm products introduced since 2007 and by the single most
important product innovation. For all manufacturing firms, we learn that the single most
important new product accounts on average for 11% of the business unit sales while all products
that are new to the firm account for 22% of business unit sales. These figures are sales-weighted.
12
If we take the simple rather than sales-weighted averages, the single most important new or
significantly improved product accounts on average for 20% of the business unit sales compared
to 27% for all products that are new to the firm. The average across firms of the ratio of the
percentage sales of the most important new product to that of all new products is 69%,
indicating, consistent with Scherer and Harhoff (2000), that the revenue impact of new products
tends to be highly skewed. Indeed, our data further suggests that a sizable share of our sample
firms have only one new or significantly improved product over the sample period.
Table 2 also provides the first industry-level estimates of patent propensity for product
innovation for all U.S. manufacturing firms.11
Fewer than half (42%) of the innovating firms
reported patenting their most significant innovation, with considerable variation across industries
and firms. Industries where the share of firms investing in R&D is higher than average also tend
to patent new products at higher than average rates. Nearly two thirds of the large firms patent
their most significant innovation compared to only 36% of small firms. Note that these figures
represent patenting by the innovating firm. When the innovation is based on an external
invention, we separately ask whether the inventor had a patent on the invention (see below).
A separate analysis of our data indicates that 94% of innovating (NTM) firms conduct
R&D and that 89% of internally generated innovations are from R&D (Lee, 2015). For
comparison, BRDIS 2011 shows that 94% of patented inventions came from R&D performing
firms (National Science Board, 2014). On the other hand, 55% of new-to-firm (NOSI) innovators
conduct R&D, which is to say 45% of new-to-firm innovators did not report conducting R&D
(Lee, 2015).
To assess the validity of our survey, we compare our findings regarding innovation rates
with those from other innovation surveys. The rank order correlation between our survey-based
NTM innovation rates at the industry level, and other innovation measures such as patenting and
the percentage of R&D performing firms (collected from our survey), are high, above 0.7.12
For
the cross-national comparisons, one might expect differences across otherwise comparable
national economies simply due to differences in the distribution of respondents across firm size
11
The Carnegie Mellon Survey reported the patent propensities only for R&D performing firms, not all innovating
firms (Cohen et al., 2000). Nonetheless, and despite the 15 years between the two surveys, the industry-level patent
propensities (i.e., percentage of innovations patented) are similar. 12
Our patent data were obtained from PATSTAT, from which we estimated the percent of firms in each industry
that had a patent application.
13
classes and industries and that innovation rates differ across these dimensions. Nonetheless, as
compared to 42% of our respondents that earned revenue in 2009 from (NOSI) products
introduced since 2007, the CIS in the UK reports that about 34% of manufacturing respondents
had introduced such a new product between 2006 and 2008. For Germany, 49% of
manufacturing respondents report introducing a new product. Turning to innovation (i.e., NTM),
about 38% of the NOSI respondents in our survey had introduced a product that was new to the
market as well. The comparable figure for the UK is 51% and that for Germany is 45%. Thus,
despite differences across the three countries in the rate at which manufacturing firms introduce
new products, the share of those products that are new to the market is similar. Moreover, the
overall rate of product innovation is also similar to our estimate of 16%, ranging from about 17%
for the UK to 22% for Germany.
4.2 Acquisition of inventions by innovating firms
A key distinction in this paper is between a product innovation (the introduction of a
product that is new to the market) and the invention that underlies it. In our survey, we asked our
innovating respondents if an outside source originated the associated invention. The possible
sources considered include: 1) a supplier; 2) a customer; 3) another firm in the same industry as
the respondent; 4) a consultant, commercial lab or engineering service provider; 5) an
independent inventor; and 6) a university or government lab. 13
Sources of invention
Table 3 shows the different external sources and the rates of reliance on each—overall,
by industry, and by firm size. We find that 49% of our respondents reported their most important
product innovation originated from an outside source. Customers are the single most likely
source of external invention, followed by suppliers. The mean number of sources for those
respondents that indicate an external source is 1.42, with 32% reporting more than one source.
The salience of customers as an external source of invention comports with prior
literature that suggests that customers are an important source of knowledge flows and ideas
(e.g., von Hippel, 1986, 2005, Cohen et al., 2002b; Klevorick et al., 1995). Table 3 also shows
13
Inquiring about which of these sources may have originated the invention if it came from outside the firm, the
precise wording in the questionnaire is: “Did any of the following originate this innovation, that is, create the overall
design, develop the prototype or conceptualize the technology?” The firm as a whole may use more sources if it
introduces multiple innovations.
14
that one of the most notable differences across size classes is that suppliers are a much more
likely source of invention for large firms. This makes sense given that it is in suppliers’ interests
to provide firms with new product ideas that increase input requirements, implying that suppliers
ought to pay particular attention to the largest firms among their customers.
If we classify consultants, independent inventors and universities all in the same category
of “technology specialists,” we see that together they account for 17% of externally sourced
inventions.14
For respondents reporting specialists as a source, the only noteworthy pattern is that
independent inventors are a markedly more common source for small firms. Though not reported
here, we find that there is no systematic tendency for any one source to be bundled with another,
except that, when a respondent indicates a customer to be the source of an invention, the
respondent is less likely to identify any other source.
We find that universities are the least frequently used source of inventions. This shows
the importance of distinguishing between knowledge sources and invention sources. For
example, Cohen et al. (2002b) reported that 29% of manufacturing firms’ R&D projects used
university research results, while only 8% of R&D projects made use of prototypes from
universities. This latter number is closer to our finding.
There is substantial variation across industries in the use of different sources. For
example, if we use the fraction of firms in an industry that perform R&D as a measure of an
industry’s technological intensity, we find that, with a correlation coefficient of 0.40, industry
technology intensity is positively related to innovators’ reliance upon universities. In contrast,
the correlation coefficients between technological intensity and reliance on customers and
suppliers are, respectively, -0.30 and -0.49, indicating that industries’ technology intensity is
negatively related to firms’ acquisition of inventions from sources in the value chain.
The patenting of externally acquired inventions also varies by source. For example,
although 42% of our innovations are patented, we find only 29% and 33% of these innovations
are patented when the customer and supplier, respectively, are the sources. This compares to
14
Our aggregation is consistent with Tether and Tajar (2008) and Nieto and Santamaria (2007) who group external
collaborations into vertical (suppliers and customers), competitors, and specialists (consultancies, research
organizations, and universities). Also, to test the robustness test of our grouping, in our regression results reported
below in tables 5a and 5b, where we compare the relative value of inventions from different sources, we split the
category of technology specialists into its three constituents. We cannot reject the null hypothesis of equality of
coefficient estimates in the majority of the specifications.
15
72%, 68%, and 59% when the invention is sourced from, respectively, a university, an
independent inventor, or an R&D service provider. Thus, where inventions originate from
technology specialists generally, the patent rate for the associated innovation is 65%. We
consider the implications of the differences in patent rates across sources in section 4.3 below.
Table 3: Sources of external invention, as % of innovators, by industry and firm size.
N Any source
%
Supp-lier %
Custo-mer %
Other firm in
Industry %
Consult./ Service provider
%
Indep. Inventor
%
Univ %
Tech. Specia-list %
Food & Bev 63 46 34 15 8 1 5 0 6 Textiles 33 50 32 26 4 3 6 0 9 Wood 52 52 22 27 11 14 1 0 15 Chemicals 102 49 17 15 5 10 3 5 16 Pharma 30 50 2 9 17 6 6 19 30 Plastics 74 53 11 28 5 11 16 4 27 Minerals 36 49 6 23 3 8 12 10 27 Metals 44 49 29 30 11 11 4 7 13 Fab’d Metal 60 48 10 38 6 0 4 3 7 Machinery 98 49 7 36 10 12 7 6 21 Electronics 50 45 11 17 10 8 6 5 14 Semicond. 91 62 16 49 9 13 8 9 23 Instruments 53 48 5 26 7 11 9 1 19 Elect Equip 98 44 12 26 4 8 7 4 17 Auto 101 52 11 28 12 6 17 15 25 Med Equip 36 49 18 22 4 13 9 15 32 Misc. 106 46 8 20 13 10 9 2 18 All Mfg. 1127 49 14 27 8 8 7 5 17 Large 457 51 22 25 9 7 4 6 14 Med 224 48 12 26 9 8 4 5 16 Small 446 49 13 28 8 8 9 5 18
Our data also indicate the extent to which startups may be a source of inventions for firms
that rely on external sources. On average, across all sources, 13% of those firms relying upon an
outside source for their invention report that the source is a startup (defined in our questionnaire
as a “new, small company”). Our aggregate figure of 13% for the contribution of startups to
other firms’ innovative activities is striking when compared to the incidence of what we define to
be startups in our manufacturing sample (i.e., single-product firms that are less than five years
old), which is 2.5%. Although one must be cautious in comparing these numbers since what
16
respondents report to be a startup and our definition of a startup may not correspond.
Nonetheless, if one assumes that these numbers are at least roughly comparable, then the large
difference between them suggest that startups may play a disproportionately important role in the
division of innovative labor.
It has been argued that patents facilitate the transfer of technology (Arora and
Ceccagnoli, 2006; Gans et al., 2008). Respondents reported that 24% of inventions acquired
from the outside were patented by the source. Inventions originating from independent inventors
were most frequently patented, at a rate of 56%, with universities next at 36%, suppliers at 34%
and R&D service providers at 28%. Inventions sourced from technology specialists (i.e.,
universities, R&D service providers and independent inventors) are patented at a rate of 37%
overall. Inventions sourced from customers are patented by the customers at a rate of 16%. In
unreported results, we find that these patterns survive when we control for industry effects and
characteristics of the innovator, such as age and size.
Channels
We asked those firms reporting acquisition of their invention from an outside source
about the channel through which they acquired that invention. The channels considered include:
1) merger, acquisition or equity purchase; 2) joint venture or cooperative R&D; 3) license; 4) a
service contract or consulting; or 5) informal means, such as informal interaction, reverse
engineering or hiring. As with sources, respondents could indicate more than one channel for
acquiring the invention from the outside.15
The mean number of channels for respondents that
used an external channel is 1.44 (with 33% of those listing a channel giving more than one). As
with sources, we found no systematic patterns in the bundling of channels, with one exception:
the use of informal channels is associated with lower than average use of other channels.
Table 4 shows that a cooperative effort (i.e., a joint venture or cooperative R&D) is the
single most important channel, accounting for 61% of externally sourced inventions in
manufacturing.16
This figure suggests that, in the majority of instances in which a firm acquires
an invention from the outside, the firm itself participates in the inventive process. Further,
15
There are 155 respondents who reported an outside source for their invention but did not identify a channel. We
treat these 155 observations as missing in our analyses based on channels. 16
Tether (2002), using the CIS2 for the UK, reports that in 1997, nearly 42% of innovators in the UK reported
cooperation with external partners, compared to about 30% in our sample.
17
suggesting that cooperative efforts may complement market channels in some cases, 27% of the
61% of firms report a cooperative effort along with a market channel (e.g., they report both a
license and joint venture or cooperative R&D). We also find, however, that 35% of respondents
report using a joint venture or cooperative R&D as their exclusive channel, with no reported use
of a license, service contract or acquisition.
Informal channels are the next most frequently cited, named by more than a third of the
respondents. A service contract or consulting is identified by a fifth of the respondents as a
channel. Licensing is named by 13% of respondents, and mergers and acquisitions are the least
frequently reported channel, though still identified by 10% of respondents acquiring their
invention from an outside source. Table 4 also shows the share of the latter three channels
combined into one category, which we call “market channels.” The column identified as
“Market” indicates that 37% of the respondents used at least one of these three channels. The
“Market only” column indicates that 16% of respondents report at least one of the three market
channels but no non-market channels (i.e., neither a cooperative effort nor an informal channel).
In other words, market channels, either alone or employed in tandem with other channels, do not
underpin the full extent of the division of innovative labor. Table 4 also shows that large firms
favor market channels relative to small firms, while small firms favor informal channels.
Medium size firms rely upon joint ventures and cooperative R&D ventures to a larger extent. We
also find that startups as a source were over-represented among those who report using market
channels. While overall 13% of the sources were reported to be startup firms, for those
innovators who used exclusively market-based channels, 23% sourced from startups.
There is significant variation in the use of channels across industries. It appears that the
more technology intensive sectors favor market channels. Indeed, one of the most R&D intensive
industries in manufacturing, pharmaceuticals, stands quite apart from almost all other industries
in its high reliance upon market channels, with 43% of the respondents reporting use of market
channels alone, particularly acquisitions and licensing. More generally, if we use the fraction of
firms in an industry that perform R&D as a measure of an industry’s technological intensity, we
find that technology intensity is positively related to the use of market channels, with a
correlation coefficient of 0.42. This suggests that the type of channels used may be related to the
nature of innovation, such as the extent to which it is science based, and therefore easier to
18
codify, or protect through patents, and, in turn, transfer across firm boundaries (cf. Arora and
Ceccagnoli, 2006; Arora and Gambardella, 1994).
Table 4: Channels for acquiring inventions, as % of innovators using external source.
N M&A JV Coop R&D
License Service Contract
Informal Market Market only
Food/Bev 26 11 73 17 20 15 32 17 Textiles 10 7 76 20 17 9 34 16 Wood 23 12 54 5 45 32 50 15 Chemicals 38 6 67 5 35 29 44 18 Pharma 13 39 40 56 8 17 82 43 Plastics 32 25 61 9 27 36 45 13 Minerals 15 13 69 12 11 56 36 17 Metals 14 19 64 3 17 46 39 18 Fab’d Metals 22 0 58 9 5 72 14 0 Machinery 35 11 53 6 17 42 34 22 Electronics 19 5 83 3 19 12 22 3 Semicond. 37 15 64 15 30 43 37 16 Instruments 19 6 41 42 14 14 60 45 Elect -Equip 27 19 62 21 36 42 55 22 Auto 37 11 67 33 18 21 53 29 Med Equip 13 15 45 17 30 31 57 24 Miscl 43 3 63 13 20 33 33 14 All Manuf 423 10 61 13 21 37 37 16 Large firms 173 18 53 21 21 29 47 24 Med firms 80 13 66 10 17 27 36 19 Small firms 170 8 60 13 23 41 36 14 Notes: Channels are not mutually exclusive. Market channels consist of licensing, contracts or M&A.
Among respondents that reported a channel, 27% of the inventions sourced externally
were patented by the source.17
Unsurprisingly, inventions sourced via licensing, or a merger and
acquisition, are most frequently reported to be patented by the source—58% for inventions
sourced via a merger or acquisition, and 76% via licensing. Also unsurprisingly, only 3% of
inventions that are sourced exclusively through informal channels are reported to be patented.
Inventions sourced exclusively via joint ventures or cooperative R&D are patented in only 15%
of the cases. These figures suggest that patents facilitate market transactions in technology. They
17
This is slightly higher than the 24% rate reported above that is obtained when we consider all respondents that use
an external invention. Not all firms reporting acquisition of their invention from the outside responded to the
question regarding channels, which accounts for the slightly different figure.
19
also suggest that patents are not common features of the division of innovative labor when non-
market channels are employed (with about three-quarters of all externally-sourced inventions not
being patented by the source).
Channels are related to sources. For example, technology specialists (i.e., R&D service
providers, universities, and independent inventors) favor market channels. In contrast, ties to
suppliers and especially customers are much less likely to involve market channels but, rather,
rely more on informal and cooperative channels, consistent with the greater trust and familiarity
bred of longstanding relationships.
Recall that Table 3 reported 49% of respondents acquiring their inventions from an
outside source. This figure includes inventions acquired via informal channels, such as reverse
engineering. This may strike some as too comprehensive. If we redefine reliance on outside
sources to consist of only acquisition of inventions via formal channels (i.e., license, merger and
acquisition, service contract, a cooperative venture), 42% of innovators rely upon an outside
source of invention. Even with this more restrictive definition, we conclude that firms’ overall
reliance upon outside sources for their most important inventions is extensive.
4.3 Relative value of inventions by source
In this section, we combine measures of value and incidence to compare the private cost
and value of inventions from different sources.18
In terms of the graphical model presented
above, we are characterizing how the cost acquiring external inventions, w, and their private
value varies across different sources. We begin by observing that it is not clear what the high
incidence of customers as a source of inventions signifies with respect to the relative value of
inventions from customers versus other sources. For example, it is possible this relatively high
incidence suggest that the preponderance of inventions drawn from customers are high value,
similar to those originating from the lead users described by von Hippel (1986). Alternatively, it
is also plausible that that the cost of acquiring inventions from customers is lower than acquiring
them from specialists because repeated interactions with customers result in greater familiarity
and trust, leading to lower search and transaction costs. Moreover, if a customer offers an
invention to its supplier, it is likely that it believes that the supplier’s adoption of that invention
18
We rely on measures of private commercial value of inventions from various sources, and infer the private costs
of obtaining inventions from these sources, as discussed below. Therefore, we can only speak of private, rather than
social, costs and benefits.
20
will be of benefit to the inventing customer firm. Accordingly, one might think that such
customer-provided inventions may well be offered at cost or even subsidized, consistent with
Harhoff et al.’s (2003) finding of pervasive “free-revealing” of inventions by customers.19
Conversely, acquiring inventions from specialists will be more costly. First, searching for
the right specialist to supply an invention may be costly, and putting in place the appropriate
contractual and legal safeguards may also result in higher costs. Second, specialists will only
benefit from the invention they provide to a focal firm through the price that they charge for the
invention. This point is also consistent with the finding that customer-sourced inventions are
patented at less than half the rate of specialist-sourced inventions—16% versus 37%. It is also
consistent with our observation that customer-sourced inventions are less likely to be conveyed
through market channels as compared to those inventions originating from technology
specialists—26% versus 62%. To address the question of the relative value of customer-sourced
versus technology specialist-sourced inventions, we will use our survey data to estimate the
relative value to the innovator of inventions from different sources.
To proceed, it is helpful to develop some notation. For the focal innovation, index the
source of invention by i, where internal invention is one possible source. The average value of an
invention from source i to an innovator is denoted by Vi. The net surplus is given by Vi – Ci,
where Ci represents the cost of acquiring and commercializing invention. The cost of “acquiring”
an internally generated invention includes the investment in research required to generate the
invention. The cost of acquiring the invention from an outside source includes any payments
made to the source, as well as any search, contracting and negotiation costs. In addition, all
inventions, whether internally generated or externally sourced, have to be developed and
commercialized. We also include costs in the relevant cost of acquisition. Value, Vi, corresponds
to what we might think of as the present value of revenues earned minus the cost of production.
The extent to which a firm’s innovation draws upon a particular source should reflect the
net surplus – the value of the invention from that source minus the cost of acquiring and
19
Harhoff et al. (2003) documented “free revealing” wherein innovative users freely reveal their invention to firms,
who can build upon it to develop a commercial product. Motivated by important examples, including open-source
software, the authors argue that a user can derive a variety of benefits from not charging manufacturers for the
invention. In Figure 1 above, ”free revealing” would shift w downward (but still would remain non-zero assuming
some search, transaction and information processing costs), thereby increasing the substitution effect (Q2 - Q1) and
increasing total innovations (Q3)
21
commercializing it. Therefore, Table 3 can be interpreted as saying that inventions sourced from
customers provide the highest net surplus compared to suppliers and technology specialists. But
is the net surplus from customers so high because the inventions from customers are really
valuable or because these inventions are easy to commercialize and can be acquired cheaply?
Our survey provides a measure of the percentage of a firm’s sales in a market generated
from their most significant innovation, which can help disentangle cost from value. Sales from
innovation have been widely used in the literature (e.g., Laursen and Salter, 2006; Frenz and
Ietto-Gillies, 2009) as a measure of the value of the innovation. As long as the sales generate
similar net margins over cost across firms in an industry, and as long as sales do not cannibalize
sales of existing products, a higher share of sales will imply a higher profitability associated with
the product. Column 1 of Table 5a presents results from regressing (OLS) the logarithm of the
percentage of sales revenue due to a focal innovation against sources of inventions. The
specification includes controls for age, size (log of employment), whether the respondent
conducts R&D, as well as 45 industry dummies at the 3 digit NAICS level of aggregation.
Internal invention is the reference source.
Assuming percentage of sales is a reasonable indicator of value, the first column in Table
5a suggests that, whereas inventions sourced from customers are less valuable than internal
inventions, those from specialists are more valuable. Note that these results are conditional upon
a source being chosen and thus may be subject to a selection bias. If the cost of inventions from
customers is less than that sourced from specialists, the marginal invention actually sourced from
a customer will also be lower in value than those from specialists. Consequently, the observed
mean value of inventions sourced from customers will also be lower than those sourced from
specialists, even if the customer and specialist-generated inventions have the same distribution of
value.
This is illustrated in Figure 2. Figure 2 depicts a hypothetical distribution of value for
different sources. Focusing on the value distribution for innovations sourced from customers (the
solid line graph), we see that the observed mean of value is equal to the mean conditional upon
the value, V, being greater than the cost, C. It is easy to see that as the cost, C, increases, the
observed mean will also increase but the probability of observing an innovation from that source
will fall. The challenge for us is to evaluate whether the value distribution of inventions from
specialists, customers and other sources differ from each other, while allowing for the costs of
22
acquisition to also differ.
Figure 2. Value and cost of acquisition of customer and specialist sourced inventions.
We take two complementary approaches. A simple test of the null hypothesis that the
value distributions are the same and only the cost of acquisition differs between customers and
specialists is to focus on the right tail of the distribution. In Figure 2, if both customers and
specialists have the distribution marked by the solid curve then the area under the curve to the
right of the point t should be the same for both customers and specialists. That is to say, the
probability of value being larger than t conditional upon the source being a customer should be
the same as that of a specialist. This is represented by the shaded triangle abt in Figure 2.
Intuitively, the right tail of the conditional distribution should be unaffected by the truncation
due to the differing costs, and thus, if the two distributions are the same, the right tail should be
the same as well.
Figure 3 shows the frequency distribution of share of sales from the focal innovation
obtained from customers (light shaded bars) and specialists (dark bars).20
Focusing on
innovations of high value (i.e., which account for 50% or more of the total sales), we see that
20
Firms indicating multiple sources were divided proportionately. Thus a respondent that indicated both a supplier
and a customer is allocated 50-50 to each source.
Value = V
Freq
uen
cy
Distribution of value of innovation conditional on the invention being acquired
ccust
Cost of acquisition
cspec
Value from specialists
Value from customers
t
a
b
23
specialists are two and half times more likely to be the source than customers, suggesting that
specialist inventions are more likely to be high value. This makes it unlikely that the difference
between customers and specialists in the conditional means of the share of sales from the focal
innovation is due entirely to differences in the costs of acquisition. Figure 3 of course does not
control for firm-characteristics or industry. Column 2 in Table 5a shows the result of a linear
probability model where the dependent variable is 1 if the % sales from the focal innovation is
greater than 50% (which corresponds to 9% of the sample of all innovations).21
Even after
controlling for business unit size, parent firm size, firm age, whether the respondent conducts
R&D or not, and industry fixed effects, we see that 17% of inventions from specialists are high
value (account for over 50% of sales) whereas only 2% of inventions from customers are high
value. This regression analysis, as well as the results in column 1 do, not, however, control for
selection.
Figure 3. Frequency distribution of customer and specialist sourced innovation by % sales
from the focal innovation.
21
Using a linear probability specification rather than a probit or logit is preferred because it is robust to
misspecification and more conservative.
80
36 43
30
149
71
42
12
0
20
40
60
80
100
120
140
160
<10% 11-25% 26-50% >50%
Fr
eq
ue
nc
y (
we
igh
ted
)
% sales in 2009 from focal innovation
Distribution of innovation value for innovations sourced from specialists and customers
Specialist
Customer
24
A different approach to the selection problem from our graphical analysis in Figure 3
incorporates the possibility that a given firm can potentially choose from among different
sources. Therefore, the observed innovation must not only cover the cost of acquiring it, but the
net surplus from obtaining it must exceed the net surplus from alternatives. As explained in
Appendix B, we follow Dubin and McFadden (1984) and use a multinomial logit framework to
express the mean value of an invention conditional upon a specific source of the invention as the
unconditional mean value of inventions from that source plus natural logarithm of the share of
the source in the industry. The latter is analogous to the familiar Heckman selection term. We
then assume that the mean value of invention from the source i in industry k for innovator j is a
linear function of a vector of firm characteristics, Xi, an industry fixed effect, zk and a source
fixed effect, sj. This leads to the specifications in columns 3 and 4 of Table 5a.
Table 5a: Value of inventions by source: % business unit sales from focal invention (The reference category for all specifications is internal invention.). OLS regressions.
Log of % BU sales from focal innovation (1)
(% BU sales from focal innovation greater than 50%) = 1 (2)
Log of % BU sales from focal innovation (Sample
Selection
correction)
(3)
(% BU sales from focal innovation greater than 50%) = 1 (Sample
Selection
correction)
(4) Customer -0.17** (0.07) -0.06*** (0.02) -0.19** (0.07) -0.07*** (0.02)
Supplier 0.07 (0.09) 0.06** (0.03) 0.03 (0.10) 0.05* (0.03)
Other Firm -0.09 (0.11) -0.03 (0.03) -0.13 (0.12) -0.04 (0.04)
Specialists 0.36*** (0.09) 0.07*** (0.03) 0.34*** (0.09) 0.07** (0.03)
Ln (BU Empl) -0.32*** (0.06) -0.07*** (0.02) -0.32*** (0.06) -0.07*** (0.02)
R&D 0.43*** (0.14) 0.09** (0.04) 0.43*** (0.14) 0.09** (0.04)
Ind. FE’s (45) Yes Yes Yes Yes
Controls Parent size, Age Parent size, Age Parent size, Age Parent size, Age
Ln (share of source)
-0.05 (0.07) -0.01 (0.02)
N 927 927 927 927
R2 0.19 0.15 0.19 0.15
Notes: The dependent variable in Columns 1 & 3 is the % sales from focal innovation. The dependent variable in
Columns 2 and 4 is a binary variable taking value 1 if the sales from the focal innovation exceed 50% of total firm
sales, and 0 otherwise. Standard errors in parentheses. Standard errors are in parentheses. *p<.10, **p<.05,
***p<.01
25
Comparing columns 1 with 3 and 2 with 4 in Table 5a shows that this correction for
selection leaves the basic pattern of results unchanged. The correction term, the log of the share
of the source, is imprecisely estimated. We obtain similar results (not reported here) when we
use a second order polynomial in the predicted probability instead of its natural log to control for
sample selection, although the coefficient on the correction term is again not significantly
different from zero.22
In sum, our results indicate that customer sourced inventions are cheaper to
acquire and commercialize and are also lower in their commercial value.
Table 5b: Value of inventions by source: Other indicators (The reference category for all specifications is internal invention.) OLS regression, all with sample selection correction.
Innovator invests equip or skills = 1 (1)
Innovator invests in sales channel =1 (2)
Innovator patents focal innovation=1 (3)
Innovator increased market share=1 (4)
Customer -0.04 (0.04) 0.00 (0.04) -0.08** (0.03) -0.05 (0.04)
Supplier -0.19*** (0.05) 0.01 (0.05) -0.11** (0.05) 0.07 (0.05)
Other Firm 0.00 (0.06) 0.02 (0.06) -0.08 (0.06) -0.06 (0.06)
Specialists 0.14*** (0.05) 0.10** (0.04) 0.28*** (0.04) 0.14*** (0.05)
Ln (Empl) 0.06* (0.03) 0.05* (0.03) 0.11*** (0.03) 0.03 (0.03)
R&D -0.02 (0.07) -0.03 (0.07) 0.18*** (0.06) 0.16** (0.07)
Ln (share of source)
-0.08** (0.03) -0.02 (0.03) 0.03 (0.03) -0.03 (0.03)
Ind. FE’s (45) Yes Yes Yes Yes
N 1012 1017 1019 916
R2 0.14 0.16 0.26 0.13
Notes: All specifications include a sample selection correction and control for parent size and age. Standard errors are in parentheses. *p<.10, **p<.05, ***p<.01
Table 5b reports the results using other indicators of value to probe the robustness of our
results on the value offered from different sources. Each of the indicators is imperfect but they
all point in the same direction, and it is the robustness of the patterns below to which we appeal.
Investment associated with the commercialization of the invention is another possible indicator
of value. We asked innovating firms whether, to commercialize the innovation, they developed
22
Dahl (2002) uses a polynomial of this share as a “correction” function to correct for the selection bias. Dahl’s
approach does not require a multinomial framework. In a multinomial framework, the natural log of the predicted
probability is the appropriate correction, as shown by Dubin and McFadden (1984).
26
new sales and distribution channels, and separately, whether they had invested in new types of
equipment or hired employees with distinct skills. A firm will undertake these actions only if it
expects a return—that is, the new product in question is valuable. In Column 1 of Table 5b, we
see that innovations sourced from specialists are associated with a greater likelihood of
investment in new plant and equipment or in hiring personnel with different expertise. By
contrast, the coefficients on inventions from other sources are not statistically different from
zero. The results in Column 2 show similarly that inventions sourced from specialists are
associated with the innovating firm investing in new sales and distribution channels, while other
sources do not have significant estimated coefficients.23
We use two other indicators of value. The first is whether the innovator filed for a patent
for some aspect of the innovation.24
Column 3 of Table 5b shows that the innovator is
significantly more likely to patent inventions sourced from specialists as compared to internally
generated inventions. Conversely, innovations sourced from customers or suppliers are
significantly less likely to be patented. Recall that these regressions included industry fixed
effects, and thus control for cross-industry differences in patent propensities.
Our final indicator considers the outcome for the firm as a whole, instead of focusing
only on the focal innovation. We asked all firms whether their market share had increased,
decreased or stayed the same over the sample period. Just over 60% of the innovating firms
report an increase in market share. In Column 4, we see that, relative to internally innovating
firms, customer-sourced innovations are associated with a higher probability of a decline in the
firm’s market share. In contrast, specialist-sourced innovations are associated with an expected
increase in the firm’s overall market share (relative to internally generated inventions).
A consistent story emerges from all the results presented in Tables 5a and 5b: the average
(unconditional) value of inventions sourced from customers are lower than those sourced from
technology specialists.25
Inventions from suppliers and other firms appear to be similar to
23
Absence of investment in sales channels and distribution may be expected in the case of customers simply
because the focal firm was already selling to the customer that was the source of the invention. But this logic does
not apply to investment in plant and new types of personnel, as considered in column 1. 24
Patents are an imperfect measure of value. An innovation may not be patentable or there may be strategic reasons
not to patent. 25
Consistent with our findings, Sofka and Grimpe (2009) report for a sample of 5000 Western European firms, the
share of revenue from innovative products is higher for firms that stress collaboration with universities and research
centers, and for firms that stress collaboration with suppliers.
27
internal invention in value. Recall, however, that Table 3 shows that, after internal invention, the
relative incidence of customers as a source of invention is the highest, followed by technical
specialists, suppliers, and other firms—and this relative incidence indicates that, on average, the
net surplus (i.e., V – C), among external sources, is highest for customers. These two conclusions
can be reconciled in only one way: while the highest gross value is realized from technology
specialists, with customer-sourced inventions having the lowest, the cost of acquiring inventions
from customers is low enough to make the net surplus—and hence, the incidence—of customer-
sourced inventions rank highest among all externally sourced inventions. In other words, the
high incidence of inventions sourced from customers reflects the low cost of acquiring and
commercializing such inventions, rather than high value.
Our results suggesting that customers offer relatively low value inventions contrast with
what we might expect if we think that the typical customers that provide inventions to firms are
the lead users described by von Hippel (1986, 2005). Though our results our entirely consistent
with a fraction of customers providing high-value inventions, they also suggest that, on average,
customer-sourced inventions will be of lower commercial value relative to those of specialists.26
And this is not surprising. First, customers may tend to anchor their suggestions on existing
products. Second, insofar as customers are other firms, Christensen’s (1997) work would suggest
that industrial customers tend to push for more incremental invention to avoid the costs of the
changing of equipment, personnel or even organizational structure that more significant
innovation on the part of the supplier may entail. And it is indeed the case that industrial
customers are more likely to be sources of invention than final consumers: in industries that the
U.S. BEA identifies as intermediate capital goods industries, 30% of our sample innovating firms
list customers as a source for their inventions, versus 18% of the innovating firms in industries
identified as final goods industries.27
In contrast to customers, technology specialists should have
no systematic interest in promoting incremental invention but rather a strong interest in
promoting more significant, valuable innovations if they are to compete against the firm’s own
R&D operations and other sources of invention.
26
Given our focus on private value, this is consistent with Riggs and von Hippel’s (1994) findings in the case of
scientific instruments that the commercial value of user-originated inventions are less than that of manufacturer-
originated innovations, even while the scientific or social value of the former may be greater. . 27
Consistent with these findings, Adams et al. (2013) highlight the important contribution of industrial users to
innovation in the semiconductor industry.
28
Our analysis not only implies that inventions from customers will be of lower average
value, but also that the cost of their acquisition and commercialization will be lower still. This
inference is consistent with our argument that customer-sourced inventions will tend to be more
incremental, and thus easier to implement and commercialize, and requiring less development
effort. Further, customer-sourced inventions will often require less investment in sales and
marketing because the customer that originated the invention is likely to be among the early
buyers. Finally, cost will be lower because it is often in the customer’s interest to charge the
upstream firm little—per Harhoff et al.’s (2003) observation of free-revealing—since the
customer benefits from the downstream firm’s commercialization of the underlying invention.
4.4 External supply and the rate of innovation
Given that sources differ in their costs and benefits, what can we infer about their
possible contributions to innovation? To address this question, we first employ the multinomial
logit framework to examine the relationship between source selection and the overall rate of
innovation. We will follow this analysis with a complementary approach to the question that
draws more directly from our survey data.
Subordinating the details of the derivation to Appendix B, if Pi is the probability of
obtaining invention from source i, and P0 is the probability of no innovation, the contribution of
source i to the overall rate of innovation in the economy – the reduction in innovation rates if that
source of invention became unavailable—is:
[P0 /(1-P0)][(Pi)/ (1-Pi)]. (1)
We can calculate the contribution of various external sources to innovation by applying
(1) to data grouped at the industry level. We use the share of inventions from source i to compute
Pi. We use the 17 industry groups used in Table 2, and take a sales-weighted average over the
industries to obtain the contribution of each source. We report bootstrap standard errors based on
500 draws. The results are summarized in Table 6.
Table 6 reports the contribution to the innovation rate for the manufacturing sector as a
whole by each source, and by external sources overall. From Table 2, we see that 16% of the
firms innovate, and in Table 3 we show that about 49% of these innovators rely upon external
sources of invention. Equation (1) implies that removing all external sources of invention would
reduce overall innovation by 45% (Table 6). Put differently, if external sources of innovation
29
were not available, only about 9% of the firms would innovate instead of 16%. Such a
hypothetical corporate self-sufficiency in innovation would cost the U.S. manufacturing dearly.
Among the external sources, we also observe, consistent with Table 3 above, that customers make
the greatest contribution, followed by specialists, suppliers, and other firms.
TABLE 6: Contribution to innovation by external source (% reduction in innovation rate if the source were not available).
Source Contribution
Customer 18.9% (1.44)
Supplier 8.5% (1.06)
Other Firm 4.6% (0.69)
Specialist 10.3% (0.95)
All external 44.6% (2.06)
Internal 47.2% (2.07) Notes: Bootstrap standard errors in parentheses based on 500 draws. Contribution to innovation calculated as: [P0 /(1-P0)][(Pi)/ (1-Pi)], where 1-P0 = % innovators, Pi = Share of inventions from source i.
These calculations are subject to the strong assumptions of the multinomial logit
framework. For one, these inferences depend upon the pattern of substitutions across sources.28
Substitution between sources means that the drop in innovation from losing a given source i is
less than Pi. However, because some of the substitution would result in the firm choosing not to
innovate, the overall rate of innovation will decline. Specifically, since over 80% of
manufacturing respondents do not innovate, this implies, in a multinomial logit framework, that
“no innovation” is the most likely counterfactual outcome if a source were removed.
Such counterfactuals are difficult to test. We have, however, another basis for projecting
what may happen in the absence of external sources of invention. For externally sourced
innovations, we asked whether the firm could have acquired the invention from another source.
Only one third (33%) of the respondents indicated that they could. Even if one assumes that the
alternative source is always internal invention, then the removal of all external sources would
imply that, at best, only a third of the 8% of respondents who sourced their inventions externally,
28
In particular, the multinomial logit setup implies, for instance, that if customers were removed as a possible
source, innovators using customer-sourced inventions would use the other sources in proportion to the shares of the
sources, so that the relative shares of the remaining sources would remain unchanged.
30
or fewer than 3%, would still innovate. If so, the rate of innovation would fall from 16% to about
11% or less. In other words, if external sources of inventions were unavailable, the overall rate of
innovation would fall by about 30%. This is a substantial decline, albeit less than the 45% figure
implied by our multinomial logit model. Thus, both of these estimates suggest a substantial
reduction in innovation if external sources of invention were not available.
Though subject to a variety of qualifications, the point of this section is to highlight that
the extent to which innovations rely upon external sources of invention and the overall rate of
innovation are jointly determined. As Figure 1 showed, external sources of invention may not
simply substitute for internal invention; they can also condition the overall rate of innovation.
Thus, removing a particular source may have two effects – a less efficient source would be
selected in some cases, and in others, the firm may not innovate at all.
5. Conclusion and implications for policy and management
Based on a survey of U.S. manufacturing firms, we find that almost half of innovating
firms had an external source for the invention that led to their most important (new to the
market) innovation. The most common external source was customers (accounting for about one
quarter of all innovations), with technology specialists (universities, consulting/contract
engineering firms, independent inventors) the second most common external source (accounting
for almost one fifth of all innovations). In addition, among these invention sources, startups
appear to be over-represented. Furthermore, we find that only a minority of these inventions was
acquired through market channels (licensing, mergers & acquisitions, and service contracts).
However, when a market-based channel was employed, the source was disproportionately likely
to have had a patent on the invention and was also disproportionately likely to be a startup (even
compared to the rate among all external sources).
Our analysis extends beyond the reporting of the findings of our survey by offering an
understanding of not only the incidence but also the relative cost and value of inventions
supplied by different sources. Using information on the share of sales accounted for by our
respondents’ most significant product innovations supplemented by data on respondents’
subsequent investments in commercialization of their innovations, patenting and market share
data, we show that, inventions sourced from technology specialists tend to be privately the most
valuable and those sourced from customers tend to provide the highest net surplus to innovators
because they are the least costly to acquire.
31
External sources are not, however, simply substitutes for internal inventions; they also
lead some firms to innovate that otherwise would not, thus augmenting the percentage of firms
that innovate. In other words, external sourcing of invention is not simply a matter of “make-or-
buy.” If firms did not have access to external sources, our data suggest that the overall rate of
innovation of 16% would likely drop to 11% or as low as 9%. While this estimate should be
interpreted with caution given the strong assumptions underlying our analyses, the main point is
that external sources contribute to the overall rate of innovation, perhaps substantially so.
Our findings raise two questions pertaining to social welfare. First, is the current state of
external sourcing socially optimal? Second, is the current balance between market and non-
market channels for acquiring inventions—where non-market channels currently predominate—
socially desirable? Our findings do not allow us to claim neither that the current state of external
sourcing of invention in U.S. manufacturing is socially suboptimal, nor whether more market-
based acquisition is indicated. We conjecture, however, that the efficiency of sourcing, and
especially market-based sourcing can be enhanced. There are a variety of policies and practices
that would make external sourcing more efficient (i.e., lowering w in Figure 1). For example,
more certainty in patent protection would likely increase licensing of inventions (Gans et al.,
2008), as would greater transparency of the terms of deals in technology markets. The supply of
high value university inventions would increase if universities were to adopt standard licensing
forms and other practices that diminish transaction costs (cf. National Research Council, 2010, p.
88). Changes in private policies and practices could also increase the efficiency of the division of
innovative labor. Increases in venture capital funding to independent inventors and startups
would improve their ability to develop ideas to the point where they can be acquired by other
firms.
There are also organizational barriers to the external sourcing of inventions. Changing the
“not-invented-here” culture, for example, that still pervades some large firms could well offer
social as well as private benefits. Indeed, our results reinforce the need to consider ways in which
managers may improve the efficiency of acquiring inventions from external sources. The
importance of listening to customers has been touted in the popular writing on the subject, and
user-innovation is by now broadly accepted by innovation scholars. Our findings suggest that
managers should also attend to high-value sources such as universities and independent
32
inventors. That independent inventors appear to disproportionately favor small firms as
compared to larger ones suggests that large firms may be missing an opportunity.29
Our findings also offer an important implication for empirical scholars studying the
determinants of R&D spending and technological change. Firms’ pervasive reliance on outside
sources for their inventions implies that innovation rates across industries depend upon factors
that condition the supply of invention from sources outside the firm and often outside the
industry. In their efforts to explain the innovation rates of firms and industries, empirical scholars
would be well advised to consider these factors along with those that are typically considered
either at the industry level, such as demand, technological opportunity and appropriability, or at
the firm level, such as firm capabilities or cash flow (cf. Cohen, 2010).
Our findings also have implications for recent debates about patent reform. Recent
debates about “patent-assertion entities,” sometimes called “patent trolls” have led to calls for
reforming the patent system to limit the ability of such entities to sue for patent infringement.
The strong link we find between patents and market-based acquisitions of inventions suggest that
patents may play an important role in facilitating market transactions for technology. We find
that a quarter of the inventions obtained from external sources were patented by the inventor,
implying that the overwhelming majority of the inventions transferred do not rely upon patents.
High value inventions from universities, independent inventors, and R&D contractors are,
however, much more likely to be patented than those from the other sources. Thus, our results
suggest that one should be quite cautious in crafting such legislation in order to not undermine
technology specialists (e.g., independent inventors, universities) who appear to rely heavily on
patent protection, and are also the most valuable source of inventions.
Our results also address important issues for survey-based indicators of innovation. A key
contribution of our study is that, by focusing questions on a single innovation, we are able to link
the particular innovation directly to several indicators of its value. This allows us to address the
ambiguity surrounding the notion of an innovation or what might be a new or significantly
improved product that has been a concern for the CIS surveys (cf. Arundel, 2006; Tether, 2001;
Smith, 2005). Our data also focus on the sources of inventions tied to specific innovations, rather
29
A different interpretation is that small firms act as a bridge between independent inventors and larger firms, who may find it too costly to deal with independent inventors. Indeed in biopharmaceuticals, biotech firms function as a bridge between university-based inventors and large pharmaceutical firms (Edwards, et al., 2006).
33
than the broader category of knowledge sources (cf. Cohen, et al., 2002b; Frenz and Ietto-Gillies,
2009), permitting us to characterize the division of innovative labor between inventors and
innovators. Also, by asking about a single innovation, we can estimate the share of innovations
that are externally acquired, thus providing precise estimates of the extent of the division of
innovative labor between inventors and innovators.
In conclusion, our results clearly suggest that the division of innovative labor involves
multiple types of actors and institutional arrangements linking them. Policy, both public and
private, must address this as a system rather than simply focus on the individual elements.
Further work on the firm and industry-level drivers of that supply, the demand, and the
transaction costs are needed to develop a system-level understanding of the division of
innovative labor.
References:
Adams, P., Fontana, R., and Malerba, F. (2013). The magnitude of innovation by demand in a sectoral system: The role of industrial users in semiconductors. Research Policy. 42, pp. 1-14.
Arora, A. & Ceccagnoli, M. (2006). Patent protection, complementary assets, and firms’ incentives for technology licensing." Management Science 52(2): 293-308.
Arora, A., & Gambardella, A. (2010). “The market for technology”, in Handbook of the
Economics of Innovation, Bronwyn H. Hall, and Nathan Rosenberg (eds.), 2010, North-
Holland, Elsevier Press, Amsterdam and Oxford
Arora, A., & Gambardella, A. (1994). The changing technology of technological change: general
and abstract knowledge and the division of innovative labour. Research Policy, 23(5),
523-532.
Arora, A., and Gambardella, A. (1990). "Complementarities And External Linkages: The
strategies of large corporations in biotechnology", Journal of Industrial Economics,
38(4): 361-379.
Arora, A., Fosfuri, A., & Gambardella, A. (2001). Markets for Technology: The Economics of
Innovation and Corporate Strategy, MIT Press, Boston MA.
Arundel, A. (2006). “Innovation survey indicators: Any progress since 1996?,” UNU-Merit
working paper.
Belderbos, R., Carree, M. and Lokshin, B. (2004). "Cooperative R&D and firm performance." Research Policy 33.10: 1477-1492.
Chesbrough, H. W. (2003). Open innovation: The new imperative for creating and profiting from
technology. Harvard Business Press.
Christensen, C. (1997). The innovator's dilemma: when new technologies cause great firms to
fail. Harvard Business Press.
Cohen, W. M. (2010). Fifty years of empirical studies of innovative activity and performance.
Handbook of the Economics of Innovation, 1, 129-213.
34
Cohen, W. M. (1995). ‘Empirical Studies of Innovative Activity,’ in Stoneman, P. (ed.),
Handbook of the Economics of Innovation and Technological Change, Basil Blackwell,
Oxford.
Cohen, W. M., Goto, A., Nagata, A., Nelson, R. R., & Walsh, J. P. (2002a). R&D spillovers,
patents and the incentives to innovate in Japan and the United States. Research
Policy, 31(8), 1349-1367.
Cohen, W. M., & Klepper, S. (1996). A reprise of size and R & D. The Economic Journal, 925-
951.
Cohen, W. M., Nelson, R. R., & Walsh, J. P. (2000). Protecting their intellectual assets:
Appropriability conditions and why US manufacturing firms patent (or not) (No. w7552).
National Bureau of Economic Research.
Cohen, W. M., Nelson, R. R., & Walsh, J. P. (2002b). Links and impacts: the influence of public
research on industrial R&D. Management Science, 48(1), 1-23.
Dahl, G. B. (2002). Mobility and the returns to education: testing a Roy Model with multiple
markets. Econometrica 70: 2367–2420.
Dubin, J., and D. McFadden (1984). “An Econometric Analysis of Residential Electric
Appliance Holdings and Consumption,” Econometrica, 52, 345–362.
Dubin, J. (1985). Consumer Durable Choice and the Demand for Electricity. New York-
Amsterdam: North-Holland Publishing Company, 1985
Edwards, M., Murray, F. & Yu, R. (2006). Gold in the ivory tower: equity rewards of
outlicensing. Nature Biotechnology 24(5): 509-515.
Frenz, M., Ietto-Gillies, G. (2009). The impact on innovation performance of different sources of knowledge: Evidence from the UK Community Innovation Survey. Research Policy 38, 1125–1135.
Freeman, C. (1991). Networks of innovators: a synthesis of research issues. Research Policy, 20(5), 499-514.
Gans, J. S., Hsu, D.H. & Stern, S. (2008). The impact of uncertainty intellectual property rights on the market for ideas: evidence from patent grant delays. Management Science 54(5): 982-997.
Griliches, Z. (1992). The Search for R&D Spillovers. The Scandinavian Journal of Economics. S29-S47.
Hagedoorn, J. (2003). Sharing intellectual property rights: an exploratory study of joint patenting
amongst companies. Industrial and Corporate Change. 12 (5): 1035-1050.
Harhoff, D., Henkel, J., von Hippel, E. (2003). Profiting from voluntary information spillovers: how users benefit by freely revealing their innovations. Research Policy 32, pp. 1753-1769.
Henderson, R., Jaffe, A., Trajtenberg, M. (1998). Universities as a Source of Commercial
Technology: A Detailed Analysis of University Patenting, 1965–1988. Review of
Economics and Statistics. February Vol. 80, No. 1, Pages 119-127
Jewkes, J., Sawers, D., & Stillerman, R. (1969). The Sources of Invention (Vol. 214, p. 215).
London: Macmillan.
Kalton, G. (Ed.) (1983). Introduction to survey sampling (Vol. 7, No. 35). SAGE Publications,
Incorporated.
Kalton, N. J. (1984). An F-space sampler (Vol. 89). CUP Archive.
35
Klevorick, A. K., Levin, R. C., Nelson, R. R., & Winter, S. G. (1995). On the sources and significance of interindustry differences in technological opportunities. Research Policy, 24(2), 185-205.
Kline, S. J., & Rosenberg, N. (1986). “An overview of innovation.” in R. Landau and N. Rosenberg, Eds. The Positive Sum Strategy: Harnessing technology for economic
growth, 14, 640. Lee, You-Na (2015). Evaluating and extending innovation indicators for innovation policy.
Research Evaluation. Published online July 30, 2015, doi:10.1093/reseval/rvv017 Laursen, K., and Salter. A. (2004). Searching high and low: what types of firms use universities
as a source of innovation? Research Policy. 33(8), 1201–1215 Laursen, K., and Salter. A. (2006). Open for innovation: the role of openness in explaining
innovation performance among U.K. manufacturing firms. Strategic Management
Journal. 27(2): 131–150.
Laursen, K., & Salter, A. J. (2014). The paradox of openness: Appropriability, external search
and collaboration. Research Policy, 43(5), 867-878.
Levin, R. C., Klevorick, A. K., Nelson, R. R., Winter, S. G., (1987). Appropriating the returns
from industrial research and development. Brookings Papers on Economic
Activity, 1987(3), 783-831.
Lhuillery, S., & Pfister, E. (2009). R&D cooperation and failures in innovation projects:
Empirical evidence from French CIS data. Research Policy, 38(1), 45-57.
Mohnen, P. and Röller, R.(2005). Complementarities in innovation policy. , European Economic
Review. Volume 49, Issue 6, August 2005, Pages 1431–1450
Mowery, D. C. (1983). The relationship between intrafirm and contractual forms of industrial
research in American manufacturing, 1900–1940. Explorations in Economic
History, 20(4), 351-374.
Mueller, W.F. (1962). “The origins of the basic inventions underlying DuPont’s major product and process innovations, 1920 to 1950.” In: Nelson, R.R. (Ed.), The Rate and Direction
of Inventive Activity. Princeton University Press, Princeton, NJ. National Research Council (2010). Managing University Intellectual Property in the Public
Interest, Washington: National Academies Press.
National Science Board (2014). Science and Engineering Indicators 2014. Arlington VA:
National Science Foundation (NSB 14-01).
Nieto, M. J., & Santamaría, L. (2007). The importance of diverse collaborative networks for the
novelty of product innovation. Technovation, 27(6), 367-377.
Pavitt, K. (1984). Sectoral patterns of technical change: towards a taxonomy and a theory.
Research Policy 13 (6), 343-373.
Pisano, G. (1990) The R&D Boundaries of the Firm, Administrative Science Quarterly 35:1, pp. 153-176.
Poot, T., Faems, D., & Vanhaverbeke, W. (2009). Toward a dynamic perspective on open
innovation: A longitudinal assessment of the adoption of internal and external innovation
strategies in the Netherlands. International Journal of Innovation Management, 13(02),
177-200.
Riggs, W., & Von Hippel, E. (1994). Incentives to innovate and the sources of innovation: the
case of scientific instruments. Research Policy, 23(4), 459-469.
Scherer, M., and Harhoff, D. (2000). Technology policy for a world of skew-distributed outcomes. Research Policy 29, 559–566.
36
Scherer, F.M., Herzstein, S., Jr., Dreyfoos, A., Whitney, W., Bachmann, O., Pesek, C., Scott, C., Kelly, T., Galvin, J. (1959). Patents and the Corporation (second ed.). Privately published, Boston, MA.
Smith, K. (2005). Measuring innovation. in J. Fagerberg, D.C. Mowery and R.R. Nelson, The
Oxford Handbook of Innovation, Oxford: Oxford University Press. Sofka, W. and Grimpe, C. (2010). Specialized search and innovation performance–evidence
across Europe. R&D Management, 40(3), 310-323. Teece, D. J. (1986). Transactions cost economics and the multinational enterprise An
Assessment. Journal of Economic Behavior & Organization, 7(1), 21-45.
Tether, B. (2001). "Identifying innovation, innovators and innovative behavior: A critical
assessment of the Community Innovation Survey,” University of Manchester, CRIC
Discussion Paper No. 48.
Tether B. (2002). Who co-operates for innovation, and why: An empirical analysis. Research
Policy. Volume 31, Issue 6, August 2002, pp. 947–967.
Tether, B. S., & Tajar, A. (2008). Beyond industry–university links: Sourcing knowledge for
innovation from consultants, private research organisations and the public science-base.
Research Policy, 37(6), 1079-1095.
Veugelers. R. & Cassiman, B. (1999). Make and buy in innovation strategies: evidence from
Belgian manufacturing firms. Research Policy. 28(1): 63–80.
von Hippel, E. (1986). Lead users: a source of novel product concepts. Management
Science, 32(7), 791-805.
von Hippel, E. (1990). Task partitioning: An innovation process variable. Research
Policy, 19(5), 407-418. von Hippel, E. (2005). Democratizing Innovation. MIT Press.
Williamson, O. E. (1991). Comparative economic organization: The analysis of discrete
structural alternatives. Administrative Science Quarterly, 269-296.
37
Appendix A: Supplemental tables
Table 1A Response rates by industry Denominator Numerator
NAICS Strata sample out Responses percent %over/under
mean 311 Food Manufacturing 1188 336 28.3% -6.8%
312 Beverage and Tobacco Product
Manufacturing
264 67 25.4% -16.4%
313 Textile Mills 262 67 25.6% -15.7%
314 Textile Product Mills 289 107 37.0% 22.0%
315-6 Apparel, Leather and Allied Product
Manufacturing
403 110 27.3% -10.0%
321 Wood Product Manufacturing 243 82 33.7% 11.2%
322 Paper Manufacturing 581 165 28.4% -6.4%
323 Printing and Related Support Activities 608 197 32.4% 6.8%
324 Petroleum and Coal Products Manufacturing 257 76 29.6% -2.5%
325 Chemical Manufacturing (except
Pharmaceutical and Medicine)
1161 349 30.1% -0.9%
3254 Pharmaceutical and Medicine
Manufacturing
724 160 22.1% -27.2%
326 Plastics and Rubber Products Manufacturing 1192 370 31.0% 2.3%
327 Nonmetallic Mineral Product Manufacturing 1118 342 30.6% 0.8%
331 Primary Metal Manufacturing 971 336 34.6% 14.1%
332 Fabricated Metal Product Manufacturing 1297 466 35.9% 18.4%
333 Machinery Manufacturing 1343 466 34.7% 14.4%
334 Computer and Electronic Product
Manufacturing (except Semicon)
1213 325 26.8% -11.7%
3344 Semiconductor and Other Electronic
Component Manufacturing
1199 370 30.9% 1.7%
335 Electrical Equipment, Appliance, and
Component Manufacturing
1231 377 30.6% 0.9%
336 Transportation Equipment Manufacturing 1190 347 29.2% -3.9%
337 Furniture and Related Product
Manufacturing
895 308 34.4% 13.4%
339 Miscellaneous Manufacturing 1286 438 34.1% 12.3%
511 Publishing Industries (except Internet) 378 90 23.8% -21.5%
512 Motion Picture and Sound Recording
Industries
186 54 29.0% -4.3%
517 Telecommunications 628 149 23.7% -21.8%
518 Data Processing, Hosting and Related
Services
517 138 26.7% -12.0%
533 Lessors of Nonfinancial Intangible Assets
(except Copyrighted Works)
181 53 29.3% -3.5%
541 Professional, Scientific, and Technical
Services
1229 340 27.7% -8.8%
Total 22034 6685 30.3% 0.0%
Notes: In this paper we do not analyze data from services
38
Table A2 Response Rates by Size and Age
Denominator Numerator
Size
strata
sample out responses percent %over/under
mean F500 1569 309 19.7% -35%
Large 6435 1524 23.7% -22%
Med 2211 659 29.8% -2%
N533 117 31 26.5% -13%
Small 7727 2880 37.3% 23%
Startup 2499 844 33.8% 11%
Tiny 936 289 30.9% 2%
Tiny Startup 540 149 27.6% -9%
Total 22034 6685 30.3% 0%
Table A3.1 Distribution of respondents, by Industry. Industry Frequency Percent
3110 Food 317 5.21
3120 Beverage 62 1.02
3130 Textile Mills 41 0.67
3140 Textile Product Mills 80 1.31
3156 Apparel, Leather 100 1.64
3210 Wood Products 79 1.3
3220 Paper 129 2.12
3230 Printing 193 3.17
3240 Petroleum 48 0.79
3250 Chemicals, other 115 1.89
3251 Basic Chemicals 76 1.25
3252 Resins 36 0.59
3254 Pharmaceuticals 133 2.18
3255 Paint 42 0.69
3256 Soap 58 0.95
3260 Plastics and Rubber 350 5.75
3270 Mineral Products 339 5.57
3310 Metals, other 235 3.86
3315 Foundries 98 1.61
3320 Fabricated Metal, other 231 3.79
3323 Structural Metals 103 1.69
3327 Machine Shops 108 1.77
3330 Machinery, other 154 2.53
3331 Heavy Machinery 40 0.66
3332 Industrial Machinery 44 0.72
3333 Commercial Machinery 37 0.61
39
3334 HVAC 36 0.59
3335 Metalworking Machinery 89 1.46
3340 Electronic Equipment, other 76 1.25
3341 Computers 30 0.49
3342 Communications Equipment 51 0.84
3344 Semiconductors 315 5.17
3345 Instruments 138 2.27
3350 Electrical Equipment 326 5.35
3360 Transportation Equipment, other 106 1.74
3361 Auto 62 1.02
3363 Auto Parts 121 1.99
3364 Aerospace 73 1.2
3370 Furniture 271 4.45
3390 Miscellaneous Manufacturing 257 4.22
3391 Medical Equipment 142 2.33
5112 Software Publishers 90 1.48
5121 Motion Picture and Sound 49 0.8
5170 Telecommunications 107 1.76
5180 Data Processing 84 1.38
5410 Professional, Scientific, and
Technical Services, other
171 2.81
5413 Architectural, Engineering 138 2.27
5415 Computer Systems Design 108 1.77
Table A3.2. Distribution of respondents, by Size.
Size strata (number of
employees)
Frequency Percent
Large (over 1000) 1477 24.26
Medium (100-1000) 1079 17.72
Small (under 100) 3532 58.02
40
Appendix B: The Multinomial Logit Model Assume all firms are potential innovators, and that a firm will introduce at most one
innovation.30
Further, assume that the firm has access to both internal as well as external
inventions. The average value of an innovation to an innovator from source i is denoted by Vi,
and the costs of acquiring the invention are denoted by Ci.
Assuming that the value of the invention to the firm from source i has a firm-specific
stochastic component, ij, the net surplus from the invention uij can be written as
uij = Vi– Ci + ij ………………………………… (1)
Firms may differ, for example, in their abilities to realize value from the same innovation,
as captured by ij. The multinomial framework assumes ij has a double exponential distribution
with mean zero and variance 2
2/6 and is iid for all i and j.
The firm chooses the option that provides the highest net surplus. Since the net surplus
may be negative (i.e., it may cost more to acquire and commercialize the invention than the
additional revenue gained), it is possible that the firm will not introduce an innovation. We
therefore include “no innovation” as an option, and normalize the net surplus associated with this
option to zero. Pij, the probability that firm j obtains its innovation from source i can be expressed
as
Pij = exp(ui/)/k(exp(uk/)), (2)
where k = 1,..i, K, and K is the number of options. Notice that the choice among the options
depends upon the net surplus associated with each option. Pij is the probability that firm j
chooses source i, and therefore the share of source i is an estimate of the expected value of Pij.
The expected net value of an innovation conditional upon it being from source i is (Dubin, 1985)
E(uij | source = i) = + Vij Ci Ln(Pi), (3)
where is Euler’s constant. If Ci does not have any unobserved variation across individuals, then
it follows that
E(Vij + ij| source = i) = + Vij Ln(Pi). (3’)
30
The actual unit of observation in our study is the business unit—reflecting a firm’s activity in one market.
41
E(Vij + ij| source = i) can be interpreted as the value earned by the firm from an innovation from
source i, which we proxy by the share of sales accounted for by the focal innovation. Equation
(3’) is analogous to the familiar Heckman selection equation where the expected value of a
variable is equal to its unconditional expectation plus a selection term. This motivates the use of
the natural log of the share of source i to control for selection. Specifically, we group the data
into the 17 industry classes used in Tables 2-4, and use the share of innovators in each industry
that use inventions from a source as the predicted probability for that industry class-source pair.
We use the natural log of this predicted probability for the source actually chosen as a regressor
to control for selection. We assume Vij is a linear function of firm and industry characteristics
plus a dummy, i, that represents the particular source, i.e.,
Vij = Xij + i (4)
Combining (3’) and (4) leads to the specification in Columns 3 and 4 of Table 5a, where
is subsumed in the constant term.
We can quantify the contribution of a particular source to the overall rate of innovation in
the economy -- the reduction in innovation rates if that source of invention became unavailable.
The probability of no innovation = P0= exp(u0/)/A, where A = k(exp(uk/)). By assumption, u0
= 0. So we have P0 = 1/A. The overall rate of innovation in the economy is equal to the
probability of innovation, which is simply 1-P0.
The probability of no-innovation without, for example, specialists = P0* = 1/A*spec, where
A*spec = k(exp(uk/)) where k specialist.
The contribution of specialists to innovation is P*0- P0 = P0 (Pspec)/ (1-Pspec). Expressed as
a share of the rate of innovation we get contribution of specialists to innovation,
(P*0- P0)/(1-P0) = [P0 /(1-P0)][(Pspec)/ (1-Pspec)]. (A3)
Equation (A3) can be interpreted as the net contribution of specialists to the overall rate
of innovation. It can be readily modified to any subset of sources, including, for instance, all
external sources of inventions.
2010 AMERICAN COMPETITIVENESS SURVEY 1
ACS: 2/19/10 (DUKE VERSION)
1. Does [insert D&B primary NAICS description from COL. E] include [THREAD IN PARENT COMPANY
NAME FROM D&B SPREADSHEET, COL. D]’s main line of business for which you are responsible?
1. YES
2. NO
3. RESPONDENT DOES NOT WORK FOR COMPANY � ASK Q1A
1A. What is the name of your company? _____________________
1B. Has [NAME OF COMPANY FROM 1A] merged with or acquired [THREAD IN PARENT COMPANY
NAME FROM D&B SPREADSHEET, COL. D]?
1. YES � Continue to Q2
2. NO � Exit
77. DK � Exit
99. REF � Exit
2. IF NO TO Q1 OR YES TO Q1B: In a couple of words, what is your main line of
business?:_______________
INTERVIEWER NOTE: USE ANY NEW COMPANY NAME AND IDENTITY OF LINE OF BUSINESS PROVIDED AS
ANSWERS TO QUESTION 2 FOR THREADING BELOW.
3. In 2009, have you earned revenue from any new or significantly improved goods or services in
[THREAD INDUSTRY] introduced since 2007, where “New” means new to your firm. Also, please
exclude simple resale of goods purchased from others or purely aesthetic changes. YES NO DK
REF
4. IF YES TO 3, ASK. ELSE, GO TO 5. In 2009, what percent of your revenues in [THREAD INDUSTRY]
is from new or significantly improved goods or services introduced since 2007?_________
ONLY PROMPT CATEGORIES IF “DK” RESPONSE TO Q4
4.1. 0%
4.2. 1-5%
4.3. 6-10%
4.4. 11-25%
4.5. 26-50%
4.6. More than 50%
4.7. DK
5. In the last 3 years, have you licensed any technology or product to another party, or have you
developed any technology or innovation for another party. YES NO DK REF
2010 AMERICAN COMPETITIVENESS SURVEY 2
- IF NO to 3 (or answered “zero” to 4) & NO to 5, GO TO BACKGROUND.
- IF NO to 3 but YES TO 5, GO TO LICENSING/CONTRACT R&D SECTION.
- IF YES TO 3 and No to 5, CONTINUE TO QUESTION 6 and skip LICENSING/CONTRACT R&D
SECTION.
- If YES to 3 and YES to 5, CONTINUE TO QUESTION 6 and answer only Q26 in
LICENSING/CONTRACT R&D SECTION.
- IF DK TO 3 AND 5: Ask “Is there someone else in the company who might be more familiar with
your new products and services in [THREAD INDUSTRY]?” GET CONTACT INFORMATION,
THANK RESPONDENT AND GO BACK TO BEGINNING OF SCRIPT.
INTERVIEWER: READ THE FOLLOWING TO THE RESPONDENT
Of all the new or significantly improved products or services you brought to market in [THREAD
INDUSTRY] during the three years, 2007-2009, think of the one that accounts for the most revenue.
Please keep this innovation in mind when answering the next set of questions.
6. Did your company patent any part of this innovation? YES NO DK REF
[INTERVIEWER: IF QUERIED, “PATENT” REFERS TO WHETHER A PATENT WAS APPLIED FOR OR GRANTED]
7. Did you introduce this innovation in your industry before any other company? YES NO DK REF
- IF NO to Questions 6 AND 7 GO TO QUESTION 26;
- TREAT A “DK” OR A “REF” AS A YES FOR PURPOSE OF SKIP LOGIC
8. In what year was this innovation introduced to the market? __________
ONLY PROMPT CATEGORIES IF “DK” RESPONSE TO Q8
8.1. 2007
8.2. 2008
8.3. 2009
8.4. DK
8.5. REF
R responds with a date prior to 2007: if YES to Q5, skip to LICENSING/CONTRACT R&D SECTION;
if NO to Q5, skip to BACKGROUND SECTION.
9. To commercialize this innovation, did you:
9.1. Develop new sales and distribution channels? YES NO DK REF
9.2. Buy new types of equipment, or hire employees with skills different from existing employees?
YES NO DK REF
10. Did you license this innovation or any related technology to others? YES NO DK REF
2010 AMERICAN COMPETITIVENESS SURVEY 3
11. Approximately how many other companies in your industry have introduced or will likely
introduce innovations that compete with this innovation? ___________
ONLY PROMPT CATEGORIES IF “DK” RESPONSE TO Q11
11.1. None
11.2. 1-3
11.3. More than 3
11.4. DK
11.5. REF
12. Did any of the following originate this innovation, that is, create the overall design, develop the
prototype or conceptualize the technology?
select all that apply
i. A Supplier [ ]
ii. A Customer [ ]
iii. Another firm in your industry [ ]
iv. A consultant, commercial lab, or engineering service provider [ ]
v. An Independent inventor [ ] vi. A collaboration between your firm and others [ ]
vii. Universities or government labs [ ]
viii. The public domain, such as, publications or public meetings [ ]
ix. Other (specify)_______________________________ [ ]
x. DON’T KNOW [ ] xi. REFUSED [ ]
13. [IF YES to ANY ONE of 12.i through 12.vi or 12.ix: Was this source [OR IF MORE THAN ONE “YES” ,
SAY INSTEAD: “Were any of these sources”] a startup company, that is, a new, small company]?
YES NO DK REF
14. [IF YES to ANY ONE of 12.i through 12.vii, or 12.ix] Did this source [OR IF MORE THAN ONE “YES”,
SAY INSTEAD: “Did any of these sources”] have a patent on any part of this innovation? YES
NO DK REF
15. [IF ANY OF 12.i THROUGH 12.ix SELECTED] How did you acquire this innovation?
select all that apply
i. Merger, acquisition or equity purchase [ ]
ii. Joint Venture or Cooperative R&D [ ]
iii. License [ ]
iv. Service contract or consulting [ ]
v. Informal means, such as reverse engineering, hiring or
informal interaction [ ]
vi. Other (specify)_______________________________ [ ]
vii. DON’T KNOW [ ] viii. REFUSED [ ]
16. [IF YES TO ANY OF 12.i through 12.ix] Could you have acquired a similar innovation from
elsewhere? YES NO DK REF
2010 AMERICAN COMPETITIVENESS SURVEY 4
17. [IF NO TO ALL OF 12.i through 12.ix] Was it mainly [THREAD COMPANY] that originated this
innovation? YES NO DK REF
17.1. IF 17 is YES, or YES to 12.vi : Did this innovation largely originate from R&D activity in your
company? YES NO DK REF
17.2. [ASK OF ALL RESPONDENTS UNLESS ANSWERED “YES” TO 17.1] Does your company conduct
R&D? YES NO DK
18. In 2009, roughly what percent of your total sales in [THREAD INDUSTRY] were from this
innovation?______
ONLY PROMPT CATEGORIES IF “DK” RESPONSE TO Q18
18.1. 1-5%
18.2. 6-10%
18.3. 11-25%
18.4. 26-50%
18.5. More than 50%
18.6. DK
18.7. REF
PREAMBLE: We are now going to ask about the business as a whole
19. Is there another division or business of [THREAD COMPANY] that is one of your main customers?
YES NO DK REF
20. Is there another division or business of [THREAD COMPANY] that is an important supplier for you?
YES NO DK REF
IF YES TO 5 AND NO TO 3 ASK ALL QUESTIONS IN LICENSING AND CONTRACT R&D SECTION
IF YES TO 5 AND YES TO 3, ASK ONLY 26 IN “LICENSING AND CONTRACT” SECTION BELOW AND SKIP
THE PREAMBLE IMMEDIATELY BELOW (PRIOR TO QUESTION 21)
LICENSING AND CONTRACT R&D SECTION
PREAMBLE: Of the innovations you licensed or developed for another party in the period 2007 to 2009,
please think of the one that generated the most revenue
21. Which one of the following would best describe this innovation?
select one
i. A product or service [ ]
ii. A process (or method for delivering a service) [ ]
iii. A technology, know-how, information or software [ ]
iv. NO REVENUE GENERATING LICENSING IN 2007-2009 � SKIP TO BACKGROUND
22. Did your company patent any part of this innovation? YES NO DK REF
2010 AMERICAN COMPETITIVENESS SURVEY 5
23. Please indicate the number of buyers or licensees for this innovation to date:__________
ONLY PROMPT CATEGORIES IF “DK” RESPONSE TO Q23
23.1. 1
23.2. 2-5
23.3. 6-10
23.4. More than 10
23.5. DK
23.6. REF
24. Did you develop this innovation for a specific customer YES NO DK REF
IF RESPONDENT NEEDS CLARIFICATION: Did the customer contract for this
innovation before it was developed?
25. How was the innovation transferred? (check all that apply)
select all that apply
i. License (excluding packaged software) [ ]
ii. Service contract or consulting [ ]
iii. Merger, acquisition or equity purchase [ ]
iv. Joint or cooperative venture [ ]
v. Other (specify)_______________________________ [ ]
vi. DON’T KNOW [ ] vii. REFUSED [ ]
26. In 2009, roughly what share of [THREAD COMPANY]’s revenues in [THREAD INDUSTRY] came from
out-licensing of technology or developing technologies or innovations for another party?________
ONLY PROMPT CATEGORIES IF “DK” RESPONSE TO Q26
26.1. None
26.2. 1-10%
26.3. 11-25%
26.4. 26-50%
26.5. More than 50%
26.6. DK
26.7. REF
2010 AMERICAN COMPETITIVENESS SURVEY 6
BACKGROUND
PREAMBLE: [SKIP THIS PREAMBLE IF COMING DIRECTLY FROM Q5] The next several questions focus on
BACKGROUND information. Please provide your best estimates.
27. Approximately how many years ago did [Thread Company Name] enter [Thread industry
name]?______
ONLY PROMPT CATEGORIES IF “DK” RESPONSE TO Q27
27.1. Less than 5 years ago
27.2. 5 or more years ago
27.3. DK
27.4. REF
28. In the last 3 years, have you introduced any new or significantly improved processes for making a
product or delivering a service in [THREAD INDUSTRY]? YES NO DK REF
29. Roughly how many employees does [THREAD COMPANY] have in [THREAD INDUSTRY], including
employees at other locations worldwide?_________________
ONLY PROMPT CATEGORIES IF “DK” RESPONSE TO Q29
29.1. Less than 25
29.2. 25 to 100
29.3. 101 to 500
29.4. 501 to 1000
29.5. Between 1,001 and 10,000
29.6. Greater than 10,000
29.7. DK
29.8. REF
30. From 2008 to 2009, what was the approximate percent change in [THREAD COMPANY]’s market
share in [THREAD INDUSTRY]?_____________ DK REFUSED
ONLY PROMPT CATEGORIES IF “DK” RESPONSE TO Q30: Do you think it…
30.1. Remained about the same?
30.2. Declined?
30.3. Increased?
30.4. DK
30.5. REF
RESPONDENT INFORMATION
31. How long you have been employed by [THREAD COMPANY]?______
32. What is your current title or position?____________________________________
33. For the report benchmarking your business, what e-mail address should we send it
to?__________
